Retroviral integration: Site matters: Mechanisms and consequences of retroviral integration site selection.

Here, we review genomic target site selection during retroviral integration as a multistep process in which specific biases are introduced at each level. The first asymmetries are introduced when the virus takes a specific route into the nucleus. Next, by co-opting distinct host cofactors, the integration machinery is guided to particular chromatin contexts. As the viral integrase captures a local target nucleosome, specific contacts introduce fine-grained biases in the integration site distribution. In vivo, the established population of proviruses is subject to both positive and negative selection, thereby continuously reshaping the integration site distribution. By affecting stochastic proviral expression as well as the mutagenic potential of the virus, integration site choice may be an inherent part of the evolutionary strategies used by different retroviruses to maximise reproductive success.

extraterminal

human immunodeficiency virus type‐1

human T lymphotropic virus type 1

integrase binding domain

integrase

integrase strand transfer inhibitor

long terminal repeat

murine leukaemia virus

nuclear pore complex

prototype foamy virus

preintegration complex

superinfection resistance

target DNA

Introduction

One of the defining events in the retroviral lifecycle is insertion of the reverse transcribed viral RNA genome into a host‐cell chromosome. This step, catalysed by the viral integrase protein (IN) establishes a stably integrated version of the virus, referred to as the provirus. The provirus provides a lasting template for viral gene expression, and its activity drives production of progeny virions by the infected cell. As the provirus constitutes an integral part of the genome, its fate is intimately linked to that of the infected cell: retroviruses persist in a host for the lifetime of the infected cells or their progeny.

Retroviral integration is not a random process: different retroviral genera (alpha‐ through epsilon‐, lenti‐ and spumaretroviruses) favour distinct chromatin environments for integration. Lentiviruses, such as the Human Immunodeficiency Virus Type‐1 (HIV‐1) for instance show a preference to integrate into actively transcribed regions 1. Conversely, gammaretroviruses such as the Murine Leukaemia Virus (MLV) target active promoter‐proximal as well as distal enhancer elements 2, 3, 4, 5. The deltaretrovirus Human T Lymphotropic Virus Type 1 (HTLV‐1) integrates more frequently in or near transcriptionally active areas 6, 7. Tree‐based clustering results of retroviruses based on their integration preferences generally map well onto the retroviral phylogenetic tree 6. This deeply rooted evolutionary link suggests that integration site selection is part of the strategy employed by retroviruses to maximise their replicative fitness.

The advent of next‐generation sequencing methods to identify integration sites 8, the resolution of crystal structures of the integration machinery (the intasome complex) 9, 10 as well as the identification of various host‐encoded integration cofactors 11, 12, 13, 14 allow us to paint a detailed picture of the retroviral integration site selection process. In this review essay, we attempt to put together the pieces required to do just that. We provide a detailed view of the known determinants of the integration site distribution and argue that target site selection can indeed be subject to evolutionary optimisation through natural selection. Finally, we discuss how the integration site distribution is further shaped by selection forces in the host. Most of the work in the field has focused on HIV and MLV owing to their relevance as a human pathogen and gene therapy vector, respectively. In the latter case, some of the implications of retroviral integration site selection have already become painfully clear. Treatment with gammaretroviral vectors in gene therapy clinical trials has led to clonal expansion and leukaemia development due to viral Long Terminal Repeat (LTR)‐driven activation of proto‐oncogenes (i.e. insertional mutagenesis) in a subset of patients 15, 16. Also in HIV‐infected patients, similar processes have recently become evident 17, 18, 19. Exploiting the extensive body of research on HIV and MLV, we will mainly discuss integration site selection as it occurs for lenti‐ and gammaretroviruses. Nevertheless, many of the mechanisms discussed here generalise to all retroviruses.

Nuclear entry route, an early determinant of viral integration site selection

The infectious cycle of retroviruses kicks off when specific receptors on a host cell are engaged by the viral envelope (Env) protein (Fig. 1A). These contacts culminate in the fusion of viral and cellular membranes and the concomitant release of the viral core into the cytoplasm. On its way towards the nucleus, the viral RNA genome is reverse transcribed into a double‐stranded DNA copy. The preintegration complex (PIC), consisting of the viral DNA associated with a specific set of viral and host proteins, then reaches the nuclear envelope. To overcome this physical barrier and gain access to the host‐cell chromatin, retroviruses have evolved different strategies. Taking distinct paths into the nucleus implies that the first chromatin environments generally encountered by various retroviral PICs are also different. If integration takes place quickly after nuclear entry, as results seem to suggest, the result is a first bias in the integration site distribution 20, 21, 22, 23.

Figure 1

Early stages of retroviral replication and nuclear entry. A: During attachment, the viral envelope engages specific host receptors leading to membrane fusion and entry of the viral core into the cytoplasm. The viral RNA genome is reverse transcribed into a dsDNA copy as the complex is trafficked towards the nucleus. Gammaretroviral preintegration complexes (PICs) wait for mitosis and nuclear envelope disassembly to access the chromatin. Lentiviral PICs traverse a nuclear pore complex (NPC). B: In gammaretroviruses, p12 stabilises the capsid and tethers the PIC to a condensed chromosome during mitosis. The complex segregates to one of the daughter cells. Upon exit from mitosis, the nuclear envelope reassembles, p12 is released and the capsid uncoats. The MLV intasome complex is set free and integration is likely to occur quickly afterwards, in the vicinity. C: In lentiviruses, nuclear import and integration are tightly coupled. The viral capsid core docks onto the NPC. Completion of reverse transcription may promote uncoating of the docked capsid. The released PIC core engages nucleoporins and import factors to gain access to the nucleoplasm. Integration occurs on a first‐come, first served basis.

Tethering of the gammaretroviral capsid to mitotic chromatin

Gammaretroviruses, such as MLV only infect dividing cells. They encode an accessory p12 protein in the group‐specific antigen (gag) gene 24. On the one hand, p12 interacts with and stabilises the capsid shell of the PIC through its N‐terminal portion, preventing untimely uncoating and ensuring completion of reverse transcription 24, 25. When the infected cell enters mitosis and the nuclear envelope breaks down, the C‐terminal region binds to a condensed chromosome, effectively tethering the capsid‐associated PIC to the chromatin (Fig. 1A and B) 26. The complex segregates along with the chromosomes to one of the daughter cells. Upon exit from mitosis, the nuclear envelope reassembles and p12 is released, orchestrating capsid uncoating 27. The MLV intasome, associated with various host and viral factors, is set free in the nucleoplasm (Fig. 1B). If integration happens shortly after the intasome first encounters chromatin, near the point of release, then site selection may be influenced by the chromatin‐binding preferences of p12. Abolishing the MLV p12‐chromatin interaction by mutagenesis blocks viral replication, and can be rescued by introducing other chromatin tethers into the mutant p12. While the preference of the rescued viruses for enhancers did not differ from wild type, some subtle variations in the distributions were observed 27. Conceivably, similar broad chromatin recognition profiles of p12 and the other tethers may hide p12's effect on site selection.

In conclusion, p12 is clearly required for integration per se. However, it does not appear to induce major biases in integration site selection, potentially owing to its broad chromatin recognition. As we will see, host cofactors are acting together with IN downstream of p12 to further pin down the gammaretroviral integration site 12, 13, 14.

Lentiviral capsid docking, uncoating and nuclear import

Lentiviruses have evolved the ability to traverse nuclear pore complexes (NPC), allowing them to infect non‐dividing, terminally differentiated cells. The viral capsid core is believed to dock onto the cytoplasmic side of the NPC through interactions with Nup358/RanBP2 (Fig. 1A and C) 28, 29. Termination of reverse transcription has been reported to promote uncoating of the docked capsid and release of the PIC 30. Through subsequent engagement of other nucleoporins such as Nup153 and Nup98‐Nup96, and import factors such as Transportin‐SR2 (TRN‐SR2, TNPO3), the released PIC gains access to the nucleoplasm 28, 31, 32.

The first nuclear subcompartment a lentiviral PIC encounters is the NPC‐associated chromatin. NPCs establish cone‐like heterochromatin‐exclusion zones at their nuclear baskets (reviewed in 33, 34, Fig. 1C). In contrast, amassment of heterochromatin into repressive lamin‐associated domains occurs adjacent to NPCs (Fig. 1C) 35.

Studies using a variety of fluorescence microscopy approaches have reported that the HIV PIC and integrated provirus preferentially localise to areas of euchromatin at the nuclear periphery 21, 22, 23. Furthermore, genes recurrently targeted for HIV integration are closely associated with NPCs 20. These results suggest that lentiviral integration site selection simply occurs in those regions of the chromatin that it encounters first as it enters the nucleus (Fig. 1C). This initial selection based on nuclear architecture may contribute to the lentiviral integration bias towards regions enriched in open chromatin marks.

The findings evince a tight coupling between lentiviral nuclear import and integration site choice. Indeed, depletion of the nuclear import factor TRN‐SR2 or NPC constituents Nup358/RanBP2, Nup153, Nup98‐Nup96 or Tpr altered the HIV integration site distribution, suggesting these proteins play a role in the nuclear import/integration pathway or in the maintenance of chromatin topology 29, 31, 36, 37, 38.

As the viral capsid affects the mode of nuclear import, it too has been shown to modulate site selection 28, 29, 36. When the entire HIV gag is replaced by its MLV counterpart, the hybrid virus integrates less into the gene‐dense environments underlying NPCs, likely reflecting an altered mode of nuclear entry 29. Moreover, several HIV capsid mutants that perturb binding to Nup358, Nup153 or cyclophilin A and/or alter capsid stability also affect integration site selection 28, 39, 40. The results further underscore the close connection between nuclear import and integration.

Host cofactors steer the PIC to specific chromatin environments

Upon entry or release of the PIC into the nucleus, IN hijacks host proteins to guide the complex to specific chromatin contexts. This likely represents an evolutionary ancient strategy as it is also adopted by LTR retrotranposons. In the case of lentiviruses, the co‐opted protein is Lens Epithelium‐Derived Growth Factor/p75 (LEDGF/p75) 11, 41. Gammaretroviral PICs on the other hand adopt Bromodomain and Extraterminal domain (BET) proteins 12, 13, 14.

LEDGF/p75 tethers lentiviral PICs to actively transcribed chromatin

LEDGF is a ubiquitously expressed chromatin reader encoded by the PSIP1 gene. It is a member of the Hepatoma‐Derived Growth Factor (HDGF) family, characterised by a conserved N‐terminal PWWP domain (named after its signature Pro‐Trp‐Trp‐Pro motif, Fig. 2A). The LEDGF PWWP specifically binds H3K36me3‐modified nucleosomes, a hallmark of active transcription 42, 43 (reviewed in 44). In contrast to the p52 splice variant, the p75 isoform harbours a C‐terminal integrase‐binding domain (IBD), which links it to a variety of cellular proteins. Aside from LEDGF/p75, only its close paralog HDGF‐Related Protein 2 (HRP‐2, HDGFRP2) carries both a PWWP and an IBD (Fig. 2A) 45.

Figure 2

Lenti‐ and gammaretroviral INs interact with specific host chromatin readers. A: Domain overview of both the p52 and p75 splice variants of H. sapiens LEDGF and HRP‐2. Abbreviations: PWWP, Trp‐Pro‐Pro‐Trp; NLS, nuclear localization signal; AT‐hook, AT‐hook minor groove DNA binding motif; SRD, supercoiled‐DNA recognition region; IBD, integrase‐binding domain. Boxes highlight, from left to right, structures of a PWWP domain bound to a methylated histone peptide (red), an AT‐hook motif (red) bound into the DNA minor groove and the IBD. B: Interaction of the LEDGF/p75 IBD (cyan) and an IN catalytic core domain dimer (green and yellow). Although the CCD of IN is essential and sufficient for interaction with LEDGF/p75, a second interface exists involving an acidic patch on the IN N‐terminal domain and a complementary basic patch on the IBD 46, 125. C: Domain overview of H. sapiens bromo‐ and extraterminal domain containing (BET) proteins. Abbreviations: Bromo, bromodomain; motif A, ∼15 aa conserved motif that may contribute to chromatin binding; NPS, N‐terminal cluster of phosphorylation sites; BID, basic residue‐enriched interaction domain; ET, extraterminal domain; SEED, Ser/Glu/Asp‐rich C‐terminal cluster of phosphorylation sites; CTM, C‐terminal motif 66, 71, 126. Boxes highlight, from left to right, the structures of a bromodomain bound to a polyacetylated histone peptide (red) and the ET domain. D: Structure of the ET domain. Residues implicated in several interaction studies with MLV IN are coloured red and can be seen lining a cleft on the domain surface. E: Sequence alignment of the C‐terminal tails of various gammaretroviral INs showing strong conservation of a ∼16 aa BET ET domain binding motif. Abbreviations: AVIRE, Avian reticuloendotheliosis virus (UniProtKB accession: P03360); MoMLV, Moloney murine leukaemia virus (P03355); AKV MLV, AKV murine leukaemia virus (P03355); MCFF MLV, Mink cell focus‐forming murine leukaemia virus (P16103); Friend MLV, Friend murine leukaemia virus (P26809); Cas‐Br‐E MLV, Cas‐Br‐E murine leukaemia virus (P08361); BAEV, Baboon endogenous virus (P10272); FEV, Feline endogenous virus (P31792); GALV, Gibbon ape leukaemia virus (P21414); WMSV, Woolly monkey sarcoma virus (P03359); KORV, Koala retrovirus (Q9TTC1).

LEDGF/p75 was first picked up as the principal cellular binding partner of HIV IN by co‐immunoprecipitation 11. Later, mutagenesis, RNA interference, transdominant overexpression of the IBD and cellular knockout studies corroborated its role in HIV‐1 replication 46, 47, 48, 49, 50, 51, 52, 53, 54. Mutagenesis studies and X‐ray crystallography were employed to unravel the molecular details of the LEDGF/p75‐IN interaction (Fig. 2B) 48, 55, 56. The main interaction interface is formed between the IBD of LEDGF/p75 and the catalytic core domain of IN. In part by binding across the IN dimer interface, LEDGF/p75 modulates IN multimerisation and activity 11.

LEDGF/p75 depletion reduces HIV‐1 integration and shifts integration preferences away from active genes 53, 57, 58. In agreement with this, mapping LEDGF/p75 chromatin occupancy revealed that LEDGF/p75 associates with active regions proportional to their transcriptional output 59. Additionally, chimeric tethers, in which the N‐terminal chromatin‐binding portion of LEDGF/p75 has been replaced by chromatin reader modules of other proteins, shifted viral integration into regions normally recognised by the other protein 60, 61, 62. In the current model, LEDGF/p75 tethers the PIC to actively transcribed chromatin, biasing integration towards these regions (Fig. 3).

Figure 3

Host cofactors steer the PIC to specific chromatin environments. Example of retroviral integration in a chromatin contact domain with an active enhancer – promoter loop driving expression of a downstream gene. Retroviral PICs hijack distinct host proteins to steer themselves into specific chromatin environments. The lentiviral PIC is tethered by LEDGF/p75 to actively transcribed regions of the chromatin decorated with H3K36me3 marks. In the case of gammaretroviruses, the intasome recruits BET proteins recognizing hyperacetylated histones H3 and H4 at active promoter‐proximal as well as distal enhancer elements.

Upon LEDGF/p75 knockout, integration into active genes still occurs more frequently than statistically expected. Part of the remaining bias is ascribed to HRP‐2 63, 64. Likely owing to its lower affinity for HIV‐1 IN and a generally lower expression level, HRP‐2 only plays a role in HIV integration site targeting when LEDGF/p75 is absent 50, 54. Nevertheless, a substantial preference for active genes remains in doubly depleted cells, supporting the involvement of additional factors such as IN itself and nuclear topology 63, 64.

BET proteins direct gammaretroviral PICs to active enhancers

When p12 orchestrates capsid disassembly upon host cell mitotic exit, the core gammaretroviral PIC is released into the nucleoplasm. The intasome then recognises its specific chromatin reader cofactors, which tether the complex to active enhancers. While an initial yeast two‐hybrid screen revealed, among many others, the BET protein BRD2 as a potential IN interaction partner 65, it was not until 2013 that three groups independently identified BET proteins as bona fide gammaretroviral integration cofactors 12, 13, 14.

The BET protein family includes BRD2, ‐3 and ‐4, which are ubiquitously expressed and BRDT, which is restricted to the testes 66 (see 67 for a recent review). BET proteins are transcriptional co‐regulators characterised by one (plants) or two (fungi/animals) bromodomains at their N‐terminus, followed by an extraterminal (ET) domain (Fig. 2C). The dual bromodomain module bestows BET proteins with a preference for hyper‐acetylated tails of histone H3 and H4 68, 69, 70. The ET domain is a protein‐protein interaction hub, connecting BET proteins to several other coactivators 67.

The interaction between BET proteins and MLV IN involves the BET ET (Fig. 2D) domain and the flexible C‐terminal tail of IN 12, 71. A ∼16 aa motif, conserved specifically among gammaretroviruses is essential and sufficient for binding (Fig. 2E) 12, 72. Specifically, alanine substitution of a single conserved Trp residue in MLV IN (W390A) was sufficient to abolish the interaction 12, 73.

Support for an integration targeting role of BET proteins came from several directions. First, the BET protein chromatin‐binding profile showed remarkable overlap with MLV integration sites 12, 13. Second, BET knockdown or specific inhibition of BET chromatin tethering through the use of bromodomain inhibitors attenuated MLV, but not HIV replication and blocked integration of MLV‐derived viral vectors 12, 13, 14. When scrutinised under these conditions, integration shifted away from enhancer elements 13. Third, a hybrid tethering factor, containing the chromatin‐binding part of LEDGF/p75 and the MLV IN‐binding part of BRD4, redirected MLV integration into active transcription units, a pattern reminiscent of HIV integration 12. Taken together, these findings firmly established BET proteins as specific mediators of integration target site selection for gammaretroviruses (Fig. 3).

Hitting the spot: The viral integrase determines the precise insertion site

LEDGF/p75 and BET proteins, respectively, guide the lenti‐ and gammaretroviral PICs to distinct chromatin contexts (Fig. 3). Subsequently, IN selects the final site as the intasome recognises a target DNA (tDNA) strand and catalyses the cutting and joining reactions that fuse viral and host genomes.

Retroviral IN contains three structurally conserved domains, connected through flexible linkers (Fig. 4A; see 74 for a recent review). X‐ray crystal structures of the spumaviral Prototype Foamy Virus (PFV) intasome complex have boosted our understanding of the retroviral integration machinery 9, 10. The intasome consists of a dimer of IN dimers assembled on the two viral DNA LTR ends (Fig. 4C). After processing the viral DNA ends, the complex captures a tDNA duplex in a groove between its inner monomers (Fig. 4C, the target capture complex, TCC). In a reaction called strand transfer, the viral DNA ends are inserted 4–6 bp apart (depending on the retroviral species) into opposing strands of the tDNA 75. The remaining single‐strand gaps are repaired by host‐cell machinery, accordingly yielding 4–6 bp duplications flanking the provirus.

Figure 4

Integrase determines the precise insertion site. A: Domain structure of HIV‐1 and MLV IN. All retroviral INs contain three structurally conserved domains connected through flexible linkers: an N‐terminal Zn2+‐binding domain, a catalytic core domain harbouring the characteristic DD‐X35‐E triad motif to chelate two catalytic Mg2+ ions, and a C‐terminal SH3‐like domain (NTD, CCD and CTD, respectively). Spuma‐ and most likely gammaretroviral INs contain an additional NTD‐extension domain (NED). B: Retroviral integration into nucleosomal DNA. The first two panels show the integration hotspots at ± 3.5 turns of DNA from the nucleosome dyad axis. The right panel suggests a general model for retroviral integration. The intasome is tethered to one or several nucleosomes through a host cofactor. Through extensive interactions with both the targeted‐ and non‐targeted DNA gyre and several core histones (potentially sensing epigenetic marks), the intasome discriminates different nucleosomes and determines the final integration site. C: Structure of the PFV intasome target capture complex (central) consisting of a dimer of IN dimers assembled on the two viral DNA LTR ends. The complex is docked onto a target DNA strand (bottom). Boxes on either side highlight direct aa – tDNA base contacts in the complex using HIV‐1 numbering. Target DNA bases are numbered starting from the sites of strand transfer on both strands. D: Sequence logos of HIV‐1 integration sites of wild type (WT) and two variant viruses (INR231G and INS119G), representing their intasome footprints on the tDNA. Arrows denote the sites of strand transfer (position 0) on the plus (black) and minus (grey) strands. A B‐DNA representation of the tDNA with the different aa‐base contacts is placed on top and positions correspond directly to those in the sequence logo. Positions where base preferences are significantly different from wild type HIV‐1 are coloured 85, 86.

Nucleosomes are the natural target for integration and induce further biases

When bound to the intasome, the tDNA is considerably kinked in order to correctly position the scissile phosphodiester bonds in the two active sites 10. This structure is in agreement with early observations that integration is favoured at inherently bendable sites, and specifically on nucleosomal DNA 76, 77. Results from cellular integration site sequencing suggest that integration is directed into outward‐facing major grooves on nucleosomal DNA 78, 79.

Recent work confirms that the PFV intasome is able to stably capture and efficiently catalyse integration into mononucleosomes 80. Integration occurs preferentially at two possible, symmetric positions, ± 3.5 turns of the DNA helix (∼36 bp) away from the nucleosome dyad axis (Fig. 4B) 80. Cryo‐electron microscopy of the intasome‐nucleosome complex revealed an extensive interaction interface 80. The contacts with the integration‐targeted gyre of the nucleosomal DNA are similar to those observed previously in the PFV intasome TCC structure (Fig. 4C) 10, 80. The second gyre is cradled by one of the two IN dimer interfaces and an inner subunit interacts with the histone H2B C‐terminal helix (Fig. 4B). Additional contacts exist between the histone H2A N‐terminal tail and the C‐terminal domain of an inner subunit of the intasome and between the C‐terminal domains of the outer subunits and other parts of the nucleosome (Fig. 4B). These contacts may allow the intasome to decode epigenetic marks 80, a situation reminiscent of the distantly related chromovirus Ty3/Gypsy LTR retrotransposons which encode a chromodomain at their IN C‐terminus, allowing them to read histone modifications and steer integration 81. Indeed, mutagenesis of residues in contact with the second gyre of DNA or with H2B reduced the PFV bias towards gene poor, lamin‐associated heterochromatin 80.

In order to avoid steric clashes, the intasome ‘lifts’ the DNA from the surface of the nucleosome (Fig. 4B). As a result, sequences yielding less stable nucleosomes appear to be better targets for capture by the intasome. Histone variants may influence integration site preferences in a similar fashion. The histone H2A L1‐loop directly underlies the integration hotspots and shows considerable variation, resulting in differential stability of nucleosomes containing certain H2A variants 80, 82. Asymmetry of the nucleosomal DNA sequence may also be expected to bias towards one of the preferred sites.

On a side note, nucleosomal capture may explain the site‐specificity of HIV‐1 integration events observed in Alu elements 17. Alu elements contribute to nucleosome positioning in the primate genome, with each dimeric Alu element having two fixed nucleosome slots 83, 84. The first of these positions a nucleosome in such a manner that the integration hotspot in the Alu element precisely overlaps one of the two hotspots on the nucleosome.

Intasome—tDNA base contacts induce integration site sequence preferences

Molecular recognition between retroviral intasome and nucleosome biases the integration site distribution. Two recent studies employed the structure of the PFV intasome TCC to predict and modify IN – tDNA contacts in the context of HIV‐1 infection (Fig. 4C) 85, 86. Two IN aa – tDNA base contacts have been discerned in the retroviral intasome, and these directly induce palindromic sequence preferences owing to the intasome central dyad axis (Fig. 4C and D) 10, 86. Using HIV‐1 numbering, the first is IN119, which projects directly into the tDNA minor groove, contacting base pairs at positions −2 and −3, as numbered from the sites of strand transfer (position 0). IN231 in HIV‐1 is the second site: it extends into the tDNA major groove and interacts with base pairs 0 and −1. IN119 is a highly polymorphic position in HIV‐1 IN that varies amongst retroviruses as well. Position IN231 is conserved in lentiviruses, but the loop in which it is located varies considerably among retroviruses. Different amino acids at these two positions yield distinct local sequence biases for viral integration 85, 86. These residues additionally modulate central tDNA bending: where interactions at IN231 can be envisioned to act as a tether, pulling the tDNA into the active sites, bulky aa at IN119 push it down at the sides, requiring stronger distortion in the centre 86.

IN119 and IN231 directly shape nucleotide preferences at tDNA positions −4 to 0 (Fig. 4D). At the site of strand transfer, retroviruses bias against T as its methyl group sterically hinders the transesterification reaction 10. Finally, as the tDNA is kinked most profoundly between the two sites of strand transfer (positions 1–3) and no aa – base contacts are established in this region, intrinsic bending potential dictates the nucleotide preferences 10, 85, 86. Depending on the size of the integration stagger, the deformation may be spread over 2–4 bp. In PFV with its 4 bp stagger, the centre of the integration site is preferentially occupied by flexible pyrimidine – purine dinucleotides 10. In the case of HIV and other retroviruses with a 5 bp stagger, minor groove compressible WWW trinucleotides such as AAA, AAT or TAA and their reverse complements dominate positions 1–3 of the integration site consensus (Fig. 4D) 86, 87.

In conclusion, capture of a target duplex by the retroviral intasome directly imposes additional restraints on the integration site, yielding fine‐grained biases in the distribution.

Local DNA contacts affect genome‐wide integration site distribution

IN – tDNA base contacts in the retroviral intasome provide mechanistic insights into the local sequence preferences for viral integration. HIV‐1 variants INS119G and INR231G however, additionally direct integration into globally less gene‐dense regions when compared to the wild‐type INS119‐R231 virus 86. The INS119G and INR231G variants exhibit wild‐type catalytic activity but alter tDNA bending requirements and consequently may reduce the shape and/or electrostatic compatibility with certain nucleosomes 86. These variants may prefer integration into the weakly bound nucleosomes present in GC‐poor regions, which are generally also gene poor. Interestingly, INS119G and INR231G viruses were associated with a more rapid disease progression in a subset of antiretroviral naive, HIV‐1 subtype C chronically infected participants of the Sinikithemba cohort 86. Additional data are required though to corroborate the observed correlation between altered intasome target site selection in these variants and viral fitness or pathogenesis.

Aside from local preferences, IN contact variants with different parts of the nucleosome can affect large‐scale biases in the integration site distribution. These global effects have now been demonstrated for HIV‐1 86 and PFV 80 and are expected to generalise to other retroviruses.

Selection pressure can directly impinge on intasome target specificity

The prevalence of IN119 and IN231 variants differs among HIV subtypes, implying that integration preferences differ slightly among subtypes 86. Additionally, HIV‐1 IN119 polymorphisms are subject to selection pressure from different sources. INS119R for instance was recently found to be an immune escape variant selected for by the HLA‐C*05 allele 88. Similarly, several HIV‐1 IN119 variants change prevalence in INSTI‐untreated versus ‐treated patients, and they have been described as secondary resistance mutations 89, 90, 91. IN119, thus, represents a position of close interaction between immune and drug selection pressure on the one hand and HIV‐1 integration site targeting on the other. While these examples are specific to HIV‐1, it is clear that any selection pressure acting on positions homologous to HIV‐1 IN119/231 in other retroviruses will influence intasome target site selection.

Proviral transcriptional activity determines the fate of the infection

At this point, the retrovirus has become an integral part of the host genome. Proviral transcriptional activity determines the ensuing course of the infection, and reshapes the integration site distribution in the population of infected cells. The activity of the viral promoter shortly after infection is probabilistically determined by a variety of cellular transcription factors binding their cognate recognition sequences in the retroviral LTR 92 (Fig. 5). Additionally, LTR activity depends on the chromatin environment 93, 94. Together, these factors result in a range of intrinsic basal transcriptional states for the provirus (Fig. 5).

Figure 5

Simplified model of HIV‐1 proviral transcription. The chromatin context, both in cis and in trans, together with the presence of various LTR‐binding transcription factors, probabilistically determine intrinsic proviral activity through various other coactivators and chromatin modifiers. When the viral trans‐activator of transcription (Tat) accumulates beyond a critical threshold, it binds the trans‐activation response (TAR) RNA stem‐loop at the 5′ end of nascent viral transcripts and recruits P‐TEFb. P‐TEFb phosphorylates RNA Pol II as well as its pausing factors enabling productive elongation. Amplification of stochastic viral expression by this Tat positive feedback creates a robust latency switch. By influencing the intrinsic LTR activity, the chromatin context and hence integration site selection influences the parameters of the switch, resulting in different probabilities for the ON and OFF states. Recognizing viral latency as a bet hedging strategy, integration site selection holds important consequences for virus evolution.

Spuma‐, lenti‐, delta‐ and epsilonretroviruses encode their own transcriptional master regulators, while alpha‐, beta‐ and gammaretroviruses rely more on host transcription factors. The HIV trans‐activator of transcription (Tat) protein amplifies stochastic expression from the LTR promoter and establishes a positive feedback loop (Fig. 5, 95, reviewed in 96). The architecture of this regulatory circuit (and perhaps those of other retroviral genera) creates a phenotypic bifurcation: the provirus can be transcriptionally active or quiescent. In the first case, progeny virions are rapidly produced, whereas in the second case, the provirus enters a potentially long‐lived latent state. Simulations of Tat protein levels in cells suggest stochastic switching between the two states (transcriptional shutdown or reactivation of the provirus) is possible under a wide array of conditions 97. The latently infected cells constitute a reservoir for the virus and represent a major hurdle towards obtaining a functional HIV‐1 cure. As will be discussed in the last section, recent reports suggest that some latently infected cells are actively proliferating and contribute to the reservoir under antiretroviral treatment 18, 19.

It is worth mentioning the phenomenon of superinfection resistance (SIR), in which an infected cell becomes resistant to superinfection by a similar type of virus. SIR has been described in several groups of viruses. As virus‐encoded proteins are usually responsible, SIR is directly related to integration and transcription in retroviruses. Often, viral (or host‐acquired) Env expression is found to interfere with infection, simply by occupying the cellular receptors for viral entry 98. The spumaviral Bet gene (Between‐env‐and‐LTR‐1‐and‐2) contributes to SIR and inhibits expression of the viral transcriptional master regulator (Tas) to maintain the original latently infected cell 98. In the case of HIV‐1, the mechanisms of SIR are not well understood 98. Nevertheless, SIR is far from absolute. Both in HIV‐1 and HTLV‐1 infected patients, evidence suggests a small percentage of the infected cells may harbour multiple proviruses 99, 100. In such cases, superinfection and viral recombination during the next round could be of evolutionary significance 101.

The chromatin environment influences proviral activity

Several studies of chromatin position effects on gene expression from the HIV‐1 LTR or active promoters showed that variability in the integration site affects transcriptional kinetics and can result in up to ∼1,000‐fold differences in expression level 93, 94, 102, 103. Besides cis‐effects, also long‐range topological interactions are known to affect proviral activity (Fig. 5) 104, 105. At least for HIV‐1 and HTLV‐1, proviral (and host gene) transcription is affected by the orientation of the integrated retrovirus through transcriptional interference 7, 106, 107, 108.

Proviruses integrated in the same genomic neighbourhood in different cells tend to have the same activation status, corroborating the existence of local features influencing latency 109. Nevertheless, the precise nature of these (epi‐) genomic features remains to be identified 108, 109, 110. For HTLV‐1, specific transcription factor binding sites and the relative position of the nearest host gene as well as its relative orientation have been linked to proviral expression 108.

The chromatin environment determines the threshold level of transcription factors required to induce LTR‐driven gene expression (Fig. 5) 111. Although the study was performed with HIV‐1, chromatin likely modulates transcription factor binding and activation in similar ways on all retroviral LTRs. Taken together, the results support the notion that intrinsic LTR activity is governed by the chromatin environment in combination with the available transcription factors (Fig. 5).

Integration site targeting is subject to natural selection: Latency and bet hedging

The Tat circuit is optimised to amplify stochastic fluctuations in gene expression 92, 95, 112. Tat feedback is sufficient to induce a robust, probabilistic latency switch independent of cellular activation state 97. In agreement with the hardwiring of this switch, the latency programme was proposed to represent a bet hedging strategy. Bet hedging is an evolutionary strategy based on stochastic phenotype switching that optimises survival in fluctuating or unfavourable environments. In the case of HIV‐1, a stochastic latency switch may allow the virus to maximise transmission by reducing extinction during mucosal infections while maintaining sufficiently high‐plasma viral loads later on 113.

While the switch itself is always present, LTR activity, determined by transcription factors and chromatin environment, directly modulates the probabilities of a provirus being in the ON or the OFF states, and hence the population of latent and productively infected cells 97. As it affects the probabilistic parameters of the latency switch, the integration site represents an inherent part of the viral bet hedging strategy and can be subject to direct evolutionary optimisation. For instance, consistent integration at permissive or repressive sites results in more proviruses in the ON or OFF state, respectively. Both scenarios are selected against due to a reduced transmissibility 113.

Not all retroviruses encode a positive transcriptional feedback loop. Nevertheless, similar bet hedging strategies may be adopted by other genera besides lentiviruses. Proviral transcription is inherently noisy, potentially magnified by the formation of a gene loop joining 5′ LTR promoter and 3′ LTR poly(A) signal 114, 115. Such loops are also observed at various cellular promoters where they impose transcriptional directionality and allow efficient recycling of the RNA Pol II machinery 115. Intrinsic transcriptional bursting from a retroviral LTR, modulated in frequency and size by the chromatin context and the presence of cognate host transcription factors, may provide enough variability to create repressed and permissive proviral states. Even without a positive feedback loop, integration site selection could thus be part of an evolutionary optimised bet hedging strategy. In this case, the phenotypic switching parameters are likely to be more dependent on host‐transcription factors and the proviral chromatin context. The MLV preference for strong enhancer elements for instance may be an adaptation to guarantee close interactions with the transcriptional machinery 4, 5.

Of note, theory predicts that modulating the latency switch to result in a latent provirus in 90–95% of infected cells would push the basic reproduction number of HIV below one, resulting in an unsustainable infection 113. This may prove to be a viable alternative to shock‐and‐kill strategies and could be envisioned using Tat antagonists 116, 117, but also with modulators of integration site targeting, chromatin or transcription factors.

The fact that LEDGF/p75 depletion hampers HIV‐1 replication 54, 64, is in line with a fitness advantage through integration site selection. However, as LEDGF/p75 affects IN oligomerisation as well as integration site targeting, it is difficult to uncouple these contributions. MLV‐derived viral vectors deficient for interaction with BET proteins suffer no or only a marginal reduction in transduction efficiency in vitro 73, 118. The effect of these mutants on multiple round viral replication (in vivo) however, remains to be determined.

Selective forces in the host shape the final integration site distribution

In vivo, selection forces continuously reshape the integration site landscape. Integration events at loci that support high expression will probabilistically lead to more active proviruses 93, 94, 109. In turn, these cells are likely to be weeded out by the immune system or die due to viral cytopathic effects 119. In the case of HTLV‐1, negative selection on the integration site distribution is dominant during chronic infection 7. Reciprocally, infected cells with a growth advantage may increase in number, for instance due to an oncogenic integration event, antigen stimulation or homeostatic proliferation 18, 19, 120, 121.

BET proteins steer gammaretroviral integration into strong promoter‐proximal as well as distal enhancer elements 4, 5. In both cases insertional mutagenesis can lead to oncogene activation and ultimately malignant transformation of the host cell 122. Insertion of an LTR in a promoter can easily be envisioned to result in transcriptional deregulation. Distal enhancers are brought into contact with promoters through DNA looping (Fig. 3) 123. Integration of an LTR enhancer in a distal regulatory region can similarly lead to activation of the target promoter via these three‐dimensional interactions, as commonly observed among oncogenic MLV integrations 122.

Recent studies have found that also HIV integration can lead to clonal expansion 17, 18, 19. HIV integration into specific cancer‐associated genes may occasionally yield a cellular survival advantage, promoting expansion and viral persistence. At least in one case a prominent clone has been found to perpetuate the infectious viral reservoir by producing replication‐competent virus in sufficient quantities to cause viraemia 18. It remains to be determined whether this finding turns out to be a general one and not all expanded clones harbour defective viruses 17, 18, 19.

Driving proliferation of the host cell may be an inherent part of the general retroviral evolutionary strategy. While some non‐acutely transforming viruses encode their own oncogenes (e.g. HTLV Tax and HBZ), others cause transformation of the host cell through insertional mutagenesis. The efficiency of cellular transformation differs considerably between retroviruses, in agreement with varying contributions to the strategies of different retroviral genera. Logically, the mutagenic potential of a retrovirus relies, at least in part, on its specific integration site biases. Integration site selection therefore contributes to this strategy as well.

Considering the potential of retroviruses to drive host cell proliferation together with stochastic viral gene expression, one may envision a reservoir of latently infected, clonally expanding cells, carrying genetically identical virus that stochastically pops up (blips). In agreement with this, genetically identical HIV variants have been reported to emerge after long‐term suppressive antiretroviral therapy 124.

Conclusions: Site matters

It has become clear that retroviral integration site selection is an intricate multistep process that starts with the mode of nuclear entry of the viral PIC (Fig. 6). Similar to the ancestral LTR retrotransposons, retroviruses co‐opt host factors to guide the viral integration machinery to specific chromatin environments (Fig. 6). Lentiviruses grab hold of the chromatin reader LEDGF/p75, directing integration into actively transcribed regions. Likewise, gammaretroviruses target enhancers by adopting BET proteins as chaperones for their integrase. Both LEDGF/p75 and BET proteins are chromatin readers, and have ample binding sites across the genome. The invading PIC therefore likely encounters an environment suitable for integration soon after it enters the nucleus. Tethered to specific chromatin regions, the intasome captures a nucleosome and catalyses the cutting and joining reactions fusing the viral and host genomes, to pin down the final integration site (Fig. 6). Direct contacts of the intasome with target DNA bases, the intrinsic bendability of the sequence, and contacts with the rest of the nucleosome further bias the integration site distribution (Fig. 6).

Figure 6

Overview of retroviral integration site targeting. Integration is believed to occur relatively quickly after the retroviral integration machinery encounters the chromatin. As a result, taking a specific route into the nucleus generates a first coarse‐grained bias in the integration site distribution (top). In this still extensive chromatin environment, further biases are introduced as the viral PIC hijacks host cofactors tethering it to well defined, functional features of the chromatin (middle). Tethered more locally to the chromatin, the intasome itself needs to capture a nucleosomal target DNA strand, catalyse the insertion process, and hence determine the final insertion points (bottom). Specific contacts between intasome and nucleosome as well as target DNA sequence considerations result in further biases in the integration site distribution. Finally, in infected hosts, selection forces continue to reshape the established population of proviruses.

Upon establishment of the provirus, the available transcription factors and the chromatin context probabilistically affect the course of the infection. Importantly, by modulating the probabilities of switching between proviral ON and OFF states, the integration site selection processes of retroviruses may be part of a bet hedging strategy and could have been optimised through natural selection. Finally, in vivo, the integration site distribution is continuously reshaped by selection processes such as immune pressure and the viral cytopathic effect, but also by clonal expansion, either passively or actively due to insertional mutagenesis (Fig. 6).